More information.
Replying to myself... this oops :

- doesn't occur up to 2.5.75-bk3 / 2.5.75-mm1
- also happens on 2.6.0-test1 :

-------------------------------------------------------------------
Unable to handle kernel NULL pointer dereference at virtual address 00000026
  printing eip:
c027014b
*pde = 00000000
Oops: 0000 [#1]
CPU:    0
EIP:    0060:[<c027014b>]    Not tainted
EFLAGS: 00210206
EIP is at __vcc_connect+0x2b/0x230
eax: 0000000a   ebx: 00000000   ecx: 00000023   edx: c02e81bc
esi: ce30d200   edi: 00000000   ebp: 00000023   esp: c8297e60
ds: 007b   es: 007b   ss: 0068
Process pppd (pid: 1097, threadinfo=c8296000 task=c8883880)
Stack: 00000000 c0306a00 c8296000 00000000 cffcf1c0 ce30d200 c9106080
c0270407
        ce30d200 00000000 00000008 00000023 00200246 0008d200 c8296000
cffcf1c0
        c8297ee8 c9106080 c026d602 c9106080 00000000 00000008 00000023
ffffffb3
Call Trace:
  [<c0270407>] vcc_connect+0xb7/0x1f0
  [<c026d602>] pvc_bind+0xb2/0x150
  [<c0212a75>] sys_connect+0x85/0xc0
  [<c026d74f>] pvc_setsockopt+0x9f/0xf0
  [<c0212f28>] sys_setsockopt+0x78/0xc0
  [<c021360d>] sys_socketcall+0xcd/0x2a0
  [<c0109087>] syscall_call+0x7/0xb
-----------------------------------------------------------------------

I'm still trying to locate the exact changeset causing this bug... but 
so far, I'm shooting in the dark : perhaps someone knowlegeable can make 
a suggestion ?
I enclose the diffstat between the two kernel trees (2.5.75-bk3 -> 
2.6.0-test1) in case it may help...

Thank you for any help,

Vincent


I wrote:
> Hi,
> 
> I get the following oops when launching ppp on 2.6.0-test1-mm1 
> (2.5.75-mm1 was fine, I've not yet tested 2.6.0-test1 itself but can do 
> it if necessary). 100% reproductible.
> I'm using ppp with pppoatm and the kernel speedtouch driver, but 
> previous versions of 2.5 were working fine with the same configuration 
> so far, so probably a recent change is to blame...
 > [...]